Unsloth ships a major update adding a browser panel beside chat, support for Google DeepMind's new EmbeddingGemma 2 multimodal embedding model, and audio tools including voice cloning. It also improves sandboxing, multi-model loading, and training memory/speed for fine-tuning workflows.
In a YouTube talk dated 2026-09-01, Google DeepMind's Nidhi Kaushik Vyas argues that commerce agents handling fuzzy intent should select each question by expected information gain rather than following a checklist. In her example, an agent asked to redo a living room on a budget asks only the highest-value question — room width — before recommending anything.
In a YouTube talk dated 2026-09-01, Google DeepMind's Nidhi Kaushik Vyas argues that commerce agents handling fuzzy intent should select each question by expected information gain rather than following a checklist. In her example, an agent asked to redo a living room on a budget asks only the highest-value question — room width — before recommending anything.
A developer released onPanda (on-Policy Alignment Data Annotator), an interactive tool for token-level visualization, editing, and control of LLM and agent outputs. Built over two years since September 2024, it targets model inspection, prompt engineering, and alignment data annotation.
A developer released onPanda (on-Policy Alignment Data Annotator), an interactive tool for token-level visualization, editing, and control of LLM and agent outputs. Built over two years since September 2024, it targets model inspection, prompt engineering, and alignment data annotation.
A Reddit user with an M2 Ultra Mac Pro (192GB unified memory) reports successfully running the Qwen3.5-397B model using a quantized GGUF (UD-Q3_K_XL) from Unsloth. They also detail contributions to llama.cpp for hybrid recurrent models, including fixing KV cache with vision and enabling slot save/restore with images.
A Reddit user with an M2 Ultra Mac Pro (192GB unified memory) reports successfully running the Qwen3.5-397B model using a quantized GGUF (UD-Q3_K_XL) from Unsloth. They also detail contributions to llama.cpp for hybrid recurrent models, including fixing KV cache with vision and enabling slot save/restore with images.
Anarlog is an open-source desktop app that records system audio, transcribes meetings on-device, and generates summaries using local models or any OpenAI-compatible provider. It solves privacy and flexibility concerns for users who want meeting notes without cloud dependency or vendor lock-in.
Anarlog is an open-source desktop app that records system audio, transcribes meetings on-device, and generates summaries using local models or any OpenAI-compatible provider. It solves privacy and flexibility concerns for users who want meeting notes without cloud dependency or vendor lock-in.
The open-source LumaBrowser released version 2.0.4, adding moss-image generation, parallel tool calls, LLM tuning, a curated Chrome store, a pseudo site API, improved text-to-speech, speech-to-text, voice cloning, and computer vision. The developer credits community feedback since the project went fully open source and points to GitHub release notes.
The open-source LumaBrowser released version 2.0.4, adding moss-image generation, parallel tool calls, LLM tuning, a curated Chrome store, a pseudo site API, improved text-to-speech, speech-to-text, voice cloning, and computer vision. The developer credits community feedback since the project went fully open source and points to GitHub release notes.